Llama 4 Scout — Spécifications
| Développeur | Meta |
|---|---|
| Type | Multimodal (MoE) |
| Modalité | Texte, image → texte |
| Paramètres | 109 milliards au total / 17 milliards actifs (MoE) |
| Fenêtre de contexte | 10 M |
| Sortie maximale | — |
| Licence | Llama 4 Community (restreint à l’UE) |
| Poids ouverts | Oui |
| Publié | 2025 |
| Prix de l’entrée | 0,10 $ / 1 million |
| Prix de la sortie | 0,30 $ / 1 million |
| Fournisseurs d'API | Meta, Together, OpenRouter |
Exécutez-le localement
| VRAM (4 bits) | ~65 Go |
|---|---|
| GPU minimal requis | H100 80 Go / Mac 128 Go |
What is Llama 4 Scout?
Llama 4 Scout is Meta’s natively multimodal open mixture-of-experts — 109B total
parameters with 17B active across 16 experts, and an industry-leading 10M-token context
window. It fits on a single 80 GB GPU at 4-bit (about 65 GB), or a 128 GB Mac, and ships
under the Llama 4 Community License with the same EU restriction as Maverick.
The 10M context is an order of magnitude beyond the 1M that counts as generous elsewhere,
and it changes what is architecturally possible: entire codebases, full document archives or
long video transcripts can go into a single prompt instead of through a retrieval pipeline.
Whether that is a good idea is a separate question — attention quality across ten million
tokens is not uniform, and retrieval still tends to beat brute force on accuracy and cost —
but for problems where chunking genuinely destroys the signal, Scout is close to unique. That
it does this while fitting on one 80 GB card is the more practical achievement. At $0.10 in /
$0.30 out per million tokens, the API is inexpensive enough to prototype against before
committing to hardware.
Llama 4 Scout pricing: API cost per 1M tokens
| Entrée (par million de jetons) | $0.100 |
|---|---|
| Sortie (par million de jetons) | $0.300 |
| Output/input ratio | 3× |
| Blended (4:1 in:out) | $0.140 per 1M tokens |
What Llama 4 Scout costs per month
Real monthly spend at a 4:1 input-to-output mix — the ratio a typical chat or RAG workload actually produces.
| Charge de travail | Jetons/mois | Cost / month |
|---|---|---|
| Projet secondaire | 1 million en entrée / 0,25 million en sortie | $0.18 |
| Petite équipe | 20 millions en entrée / 5 millions en sortie | $3.50 |
| Production | 200 millions en entrée / 50 millions en sortie | $35 |
Run your own numbers in the Calculateur de coûts des API IA.
Cheaper alternatives to Llama 4 Scout
| Modèle | Coût combiné par million de dollars | You save |
|---|---|---|
| Qwen3 32B ouverte | $0.120 | 14% cheaper |
| Gemma 3 27B ouverte | $0.0960 | 31% cheaper |
Self-host or pay the API?
Llama 4 Scout is open-weight, so you can run it yourself. It needs ~65 Go of VRAM at 4-bit (H100 80GB / Mac 128GB). Self-hosting only beats the API once your volume is high enough to keep that hardware busy — the calculateur auto-hébergement vs API works out the break-even point for your token volume.
Questions fréquemment posées
How much does Llama 4 Scout cost per 1M tokens?
Llama 4 Scout costs $0.100 per 1M input tokens and $0.300 per 1M output tokens. At a typical 4:1 input-to-output mix that blends to about $0.140 per 1M tokens.
How much does Llama 4 Scout cost per month?
A small-team workload of 20M input and 5M output tokens a month costs about $3.50 on Llama 4 Scout. A side project (1M in / 0.25M out) costs roughly $0.18.
What is a cheaper alternative to Llama 4 Scout?
Qwen3 32B is the strongest cheaper option in our database at $0.120 per 1M blended — about 14% less than Llama 4 Scout. It is also open-weight, so self-hosting is an option.
Can I run Llama 4 Scout locally?
Yes. Llama 4 Scout is open-weight and needs about ~65 GB of VRAM at 4-bit quantisation (H100 80GB / Mac 128GB).
Why does Llama 4 Scout charge more for output than input?
Output tokens are generated one at a time and cannot be batched the way a prompt can, so they cost the provider more to serve. Llama 4 Scout charges 3× more for output, which is why prompt-heavy workloads are far cheaper to run than generation-heavy ones.
Prices are the published list rates for the model's primary API and are reviewed as providers change them. Volume, batch and cached-input discounts are not included. Compare every model side by side in the Base de données des modèles d'IA ou le Classement des grands modèles linguistiques (LLM).
